Read Agent Output

When to use

You need to inspect what a session (spawned via ap-spawn-agent) actually
produced — a quick tail check, the full transcript, a cost readout, or a
human-readable summary/timeline.

MCP tools

{ "tool": "agent_output", "args": { "sessionId": "sess_5f9a3c38", "lastN": 50, "clean": true } }

lastN caps at 500. clean:true strips ANSI escape codes for readable
text.

{ "tool": "agent_export", "args": { "sessionId": "sess_5f9a3c38", "format": "markdown" } }

agent_export works on stopped sessions too — use it when you need the
whole conversation, not just the tail.

{ "tool": "conversation_read", "args": { "sessionId": "sess_5f9a3c38" } }
{ "tool": "conversation_locate", "args": { "sessionId": "sess_5f9a3c38", "query": "signup validation fix" } }
{ "tool": "session_usage", "args": { "sessionId": "sess_5f9a3c38" } }
{ "tool": "agentproto_session_story", "args": { "sessionId": "sess_5f9a3c38" } }
{ "tool": "summarize_session", "args": { "sessionId": "sess_5f9a3c38" } }

• session_usage — cost and token counts for the session.
• agentproto_session_story — a human-readable timeline of what happened.
• summarize_session — a one-sentence summary plus a coarse state (e.g.
  running / idle / done).

CLI

agentproto sessions export sess_5f9a3c38
agentproto sessions story sess_5f9a3c38
agentproto sessions mirror sess_5f9a3c38   # read-only live tail

HTTP

curl http://127.0.0.1:18790/sessions/sess_5f9a3c38/events

Gotchas

• agent_output reads from a tail ring buffer, not the full
  transcript — long sessions will have earlier output evicted. Use
  agent_export when you need everything, including the beginning.
• The awaitingInput flag over-signals: it fires both when a turn
  simply finished and when the agent is genuinely blocked waiting on a
  question. Don't trust the flag alone — read the last content line of the
  output to tell "done" from "stuck waiting on you."

Pointers

• ap-spawn-agent — where the session you're reading came from
• ap-prompt-agent — sending a follow-up once you've read the output
• supervise-long-missions — longer-running supervision patterns
• agentproto — general daemon driving reference